Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 Locally nite spaces and the join operator Erik Melin Department of Mathematics, Uppsala University, Sweden [email protected] Abstract Keywords: The importance of digital geometry in image processing is well documented. To understand global properties of digital spaces and manifolds we need a solid understanding of local properties. We shall study the join operator, which combines two topological spaces into a new space. Under the natural assumption of local niteness, we show that spaces can be uniquely decomposed as a join of indecomposable spaces. join operator, Alexandrov space, smallest-neighborhood space, locally nite space. 1. Introduction Topological properties of digital images play an important role in image processing. Much of the theoretical development has been motivated by the needs in applications, for example digital Jordan curve theorems and the theory of digitization. A classical survey is [11] by Kong and Rosenfeld. Inspired by the new mathematical objects that have emerged from this process, mathematicians have started to study digital geometry from a more theoretical perspective, developing the theories in dierent directions. Evako et al. [5, 6] considered, for example, n-dimensional digital surfaces satisfying certain axioms. These surfaces were later considered by Daragon et al. [4]. Khalimsky spaces as surfaces, embedded in spaces of higher dimension have been studied the author in [15]. We shall study, not digital spaces, but a tool that can be used in such a study: the join operator. Evako used an operation on directed graphs called the join to aid the analysis. We will generalize this construction and study its properties. In particular we show how a space can be decomposed into indecomposable pieces put together by the join operator. We will give conditions for uniqueness of this decomposition. 2. Digital spaces and the Khalimsky topology We present here a mathematical background. The purpose is primarily to introduce notation and formulate some results that we will need. A reader not familiar with these concepts is recommended to take a look at, for example, Kiselman's [9] lecture notes. 63 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 2.1 Topology and smallest-neighborhood spaces Not all topological spaces are reasonable digital spaces, but the class of nite topological spaces is too small. It does not include In every topological space, a whereas an arbitrary nite Zn . intersection of open sets is open, intersection of open sets need not be open. If the space is nite, however, there are only nitely many open sets, so nite spaces fulll a stronger requirement: arbitrary intersections of open sets are open. Alexandrov [1] considers topological spaces, nite or not, that fulll smallest-neighborhood Alexandrov spaces, but this this stronger requirement. We shall call such spaces spaces. Another name that is often used is name has one disadvantage: it has already been used for spaces appearing in dierential geometry. Let B be a subset of a topological space intersection of all closed sets containing by B. We shall instead write CX (B) B. X. The closure of B is the The closure is usually denoted for the closure of B in X. This notation allows us to specify in what space we consider the closure and is NX dened below. X as above, we dene NX (B) to be the intersection of all open sets containing B . In general NX (B) is not an open set, but in a smallest-neighborhood space it is; NX (B) is the smallest neighborhood containing B . If there is no danger of ambiguity, we will just write N (B) and C (B) instead of NX (B) and CX (B). If x is a point in X , we dene N (x) = N ({x}) and C (x) = C ({x}). Note that y ∈ N (x) if and only if x ∈ C (y). We have already seen that N (x) is the smallest neighborhood of x. also a notation dual to Using the same B and Conversely, if every point of the space has a smallest neighborhood, then an arbitrary intersection of open sets is open; hence this existence could have been used as an alternative denition of a smallest-neighborhood space. A point x point is called pure, open if N (x) = {x}, that is, if {x} is open. The C (x) = {x}. If x is either open or closed it is called called mixed. is called closed otherwise it is if Adjacency and connectedness A topological space X is called connected closed and open, are the empty set and nent if the only sets, which are both X itself. A connectivity compo- (sometimes called a connected component) of a topological space is a connected subspace which is maximal with respect to inclusion. Two distinct points x and y in X are called adjacent if the subspace {x, y} is connected. It is easy to check that x and y are adjacent if and only y ∈ N (x) or x ∈ N (y). Another equivalent condition is y ∈ N (x) ∪ C (x). The adjacency neighborhood of a point x in X is denoted ANX (x) and is the set NX (x) ∪ CX (x). It is practical also to have a notation for the set of points adjacent to a point, but not including it. Therefore the adjacency set 64 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 x, denoted AX (x), is dened to be AX (x) = ANX (x) r {x}. A (x) and AN(x). A point adjacent to x is sometimes called a neighbor of x. This terminology, however, is somewhat dangerous since a neighbor of x need not be in the smallest neighborhood of x. in X of a point Often, we just write Separation axioms Kolmogorov's separation axiom, also called the two distinct points x and y, T0 axiom, states that given there is an open set containing one of them but not the other. An equivalent formulation is that x=y for every Clearly any T1/2 x and y. The space is also T1/2 N (x) = N (y) implies axiom states that all points are pure. T0 . The next separation axiom is the T1 axiom. It states that points are closed. In a smallest-neighborhood space this implies that every set is closed and hence that every set is open. Therefore, a smallest-neighborhood space satisfying the T1 axiom must have the discrete topology, and thus, is not very interesting. Duality Since the open and closed sets in a smallest neighborhood space X satisfy exactly the same axioms, there is a complete symmetry. Instead of calling the open sets open, we may call them closed, and call the closed sets open. Then we get a new smallest-neighborhood space, called the we will denote by X 0. dual of X , which The AlexandrovBirkho preorder There is a correspondence between smallest-neighborhood spaces and par- X be a smallest-neighborhood space and dene x 4 y to hold if y ∈ N (x). We shall call this relation the Alexandrov Birkho preorder. It was studied independently by Alexandrov [1] and by tially preordered sets. Let Birkho [3]. x is x 4 x) x, y, z ∈ X , x 4 y and y 4 z imply x 4 z ). A relation satisfying these conditions is called a preorder (or quasiorder ). A preorder is an order if it in addition is anti-symmetric (for all x, y ∈ X , x 4 y and y 4 x imply x = y ). The AlexandrovBirkho preorder is an order if and only the space is T0 . In conclusion, it would therefore be possible The AlexandrovBirkho preorder is always reexive (for all and transitive (for all to formulate the results of this paper in the language of orders instead of the language of topology. 65 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 2.2 The Khalimsky topology We shall construct a connected topology on Z, which was introduced by Em Khalimsky (see Khalimsky et al. [8] and references there). Pm = ]m − 1, m + 1[, and if m is an even Pm = {m}. The family {Pm }m∈Z forms a partition of the Euclidean line R and thus we may consider the quotient space. If we identify each Pm with the integer it contains, we get the Khalimsky topology on Z. We call this space the Khalimsky line. Since R is connected, the Khalimsky If m is an odd integer, let integer, let line is a connected space. It follows readily that an even point is closed and that an odd point is open. In terms of smallest neighborhoods, we have odd and N (n) = {n − 1, n, n + 1} if n N (m) = {m} if m is is even. Perhaps this topology should instead be called the AlexandrovHopf Khalimsky topology, since it appeared in an exercise [2, I:Paragraph 1:Exercice 4]. However, it was Khalimsky who realized that this topology was useful in connection with digital geometry and studied it systematically. Since this topology is also called the Khalimsky topology in the literature, we will keep this name. A dierent approach to digital spaces, using cellular complexes, was introduced independently by Herman and Webster [7] and by Kovalevsky [13]. The results of this article apply to such spaces, since they are topologically equivalent to spaces of the type introduced in this article. See, for example, Klette [10]. Khalimsky intervals and arcs Let a and b, a 6 b, be integers. A Khalimsky interval is an interval [a, b]∩Z of integers with the topology induced from the Khalimsky line. We will denote [a, b]Z and call a and b its endpoints. A Khalimsky arc in X is a subspace that is homeomorphic to a Khalimsky If any two points in X are the endpoints of a Khalimsky arc, we X is Khalimsky arc-connected. such an interval by a topological space interval. say that Theorem 1. A T0 smallest-neighborhood space is connected if and only if it is Khalimsky arc-connected. A proof can be found in [14, Theorem 11]. Slightly weaker is [8, Theorem 3.2c]. The theorem also follows from Lemma 20(b) in [12]. length of a Khalimsky arc, A, to be the number of points L(A) = card A − 1. If a smallest-neighborhood space X Let us dene the in is A T0 minus one, and connected, Theorem 1 guarantees that length of the shortest arc connecting x y in X is a nite number. This observation ρX on X , which we call the arc metric. and dene a metric ρX (x, y) = min(L(A); A ⊂ X is a Khalimsky arc containing 66 allows us to x and y). Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 The arc metric is dened on the topology of X X, but it is important to bear in mind that is not the metric topology dened by ρX . The metric topology is of course the discrete topology. Examples of smallest-neighborhood spaces We conclude this section with a few examples to indicate that the class of smallest-neighborhood spaces and spaces based on the Khalimsky topology is suciently rich to be worth studying. Example 1. The Khalimsky plane is the Cartesian product of two KhalimKhalimsky n-space is Zn with the product topology. sky lines and in general, Points with all coordinates even are closed and points with all coordinates odd are open. Points with both even and odd coordinates are mixed. It is easy to check that Example 2. integer A (p) = {x ∈ Zn ; kp − xk∞ = 1} if p is pure. Zm = Z/mZ for some even Khalimsky circle. If m > 4, Zm is We may consider a quotient space m > 2. Such a space is called a a compact space that is locally homeomorphic to the Khalimsky line. (If m were odd, we would identify open and closed points, resulting in a space with the indiscrete or chaotic topology, i.e., where the only open sets are the empty set and the space itself.) 3. Locally nite and locally countable spaces A topological space actually stored in a computer is nite. Nevertheless, it is of theoretical importance to be able to treat innite spaces, like Z2 , since the existence of a boundary often tends to complicate matters. On the other hand, spaces where a points can have an innite number of neighbors seems less likely to appear in computer applications. In this situation, not even the local information can be stored. Denition 1. A smallest-neighborhood space is called ery point in it has a nite adjacency neighborhood. countable adjacency neighborhood, it is called locally nite if ev- If every point has a locally countable. We need to assume the axiom of choice (in fact the countable axiom of choice, which states that from a countable collection of nonempty sets we can select one element from each set, is sucient) to prove the following. Proposition 1. Let X be a locally nite space. If X is connected, then X is countable. Proof. If X is not T0 , we consider the quotient space identical neighborhoods have been identied. X̃ X̃ is T0 X̃ , where points with and X is countable if is countable. It is therefore sucient to prove the result for 67 T0 spaces. Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 x be any point in X . By induction based on local niteness, the Bn (x) = {y; ∈ X; ρX (x, y) 6 n}, where ρX is the arc-metric on X (here we use that the space is T0 ), is nite for every n ∈ N. Since X = S∞ n=0 Bn (x), the result is true, since the countable axiom of choice implies Let ball that a countable union of nite sets is countable. In a similar way, we can also characterize countable smallest-neighborhood spaces. Proposition 2. A smallest-neighborhood space X is countable if and only if it is locally countable and has countably many connectivity components. Proof. It is clear that a countable space is locally countable and has count- ably many components. The countable axiom of choice implies that countable unions of countable sets are countable. So if X is connected and locally countable, a slightly modied version of the proof of Proposition 1 shows that X is countable. If X has countably many connectivity components and each component is countable, then X is countable. The following proposition states that the set of open points is dense in a locally nite space (and by duality a corresponding result holds for the closed points). It is obvious that the open points form the smallest dense set. Proposition 3. Let X be a smallest-neighborhood space and let S ⊂ X be the set of open points in X and T be the set of closed points. If X is T0 and locally nite, then X = C (S) = N (T ). Proof. y0 X . Let Y0 = N (y0 ). If Y0 is a singleton set, then y0 ∈ S . Otherwise, for k > 0, choose yk+1 ∈ Yk r {yk } and let Yk+1 = N (yk+1 ). Clearly Yk+1 ⊂ Yk and since X is T0 , it follows that yk 6∈ Yk+1 . Repeat the construction above recursively until Yk is a singleton set, at most card(Y0 )−1 steps are needed. Note that yk is an open point and that y0 ∈ C (yk ). Since y0 was arbitrarily chosen, we are done. We will prove the st equality, the other follows by duality. Let be any point in The relation in the proposition need not hold if the space is not locally nite. Consider the space (non-trivial) open sets are given by intervals Z where the ]−∞, m]Z , m ∈ Z. This space contains no open point. −∞ and declare the open sets to m ∈ Z ∪ {−∞}, then the point −∞ is open C (−∞). Thus, nite neighborhoods are not On the other hand, if we add the point be intervals [−∞, m]Z , where and the whole space equals necessary for the conclusion of the proposition. 68 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 4. The join operator A well-known way to combine two topological spaces, the coproduct (disjoint union), X ` Y. The pieces, X and Y is to take X and Y , are com- pletely independent in this construction. We shall introduce another way of combining two spaces, namely the join operator. Denition 2. Y, denoted Let X X ∨Y, and Y be two topological spaces. The ˙ is declared Y , where a subset A ⊂ X ∪Y (i) A ∩ X is open in X and A ∩ Y = ∅, or (ii) A ∩ X = X and A ∩ Y is open in Y . and (ii) of X and X to be open if either B ⊂ X ∨ Y is closed if and only B ∩ X is closed in X and B ∩ Y = Y , or B ∩ X = ∅ and B ∩ Y is closed in Y . Note that a set (i) join is a topological space over the disjoint set union of if While this denition makes sense for any topological space, it is a strange operation on large spaces. R ∨ S 1 , is a compact cannot be T1 unless X or Y circle, For example, the join of the real line and the space, which is T0 T1 . but not In fact, X ∨Y is empty. From now on, we shall only consider the join of smallest-neighborhood X and Y X ∨ Y is given by NX∨Y (y) = NY (y) ∪ X if y ∈ Y . Ap- spaces. In this case, the denition boils down to the following. If are smallest-neighborhood spaces, then the topology of NX∨Y (x) = NX (x) if x∈X and parently, the join of two smallest-neighborhood spaces is a smallest-neighborhood space. This denition of the join is compatible with the join of directed graphs, see [6, p. 111]. In terms of the AlexandrovBirkho preorder, every element of X X in is placed on top of Y Formally, the order, i.e., here denote Ord(X ∨ Y ) is declared to be larger than any element of Y ; X ∨ Y . This motivates also our notation X ∨ Y . pairs (x, y) satisfying x < y , on X ∨ Y , which we is Ord(X ∨ Y ) = Ord(X) ∪ Ord(Y ) ∪ X × Y. The join of two connected locally nite spaces need not be locally nite. In fact, the join of two spaces is locally nite if and only if the spaces are nite and locally countable if and only if both are countable. In view of Proposition 2, the join of two connected and locally countable spaces is locally countable. When Here {p} p is a point, we write p∨X instead of {p} ∨ X . is the topological space with one point. 4.1 Basic properties The following three properties in the next proposition are straightforward to prove. 69 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 Proposition 4. The join operator has the following properties for all smallest-neighborhood spaces X, Y and Z . (i) X = ∅ ∨ X = X ∨ ∅ (has a unity). (ii) (X ∨ Y ) ∨ Z = X ∨ (Y ∨ Z) (is associative). 0 0 0 (iii) (X ∨ Y ) = Y ∨ X . The following proposition lists some topological properties. Proposition 5. Let X and Y be smallest-neighborhood spaces. (i) (ii) (iii) X ∨ Y is T0 if and only if X and Y are T0 . X ∨ Y is compact if and only if Y is compact. If X 6= ∅ and Y 6= ∅ then X ∨ Y is connected. Proof. To prove (i), note that X is open in X ∨ Y . If x ∈ X and y ∈ Y , X is an open set containing x but not y . It follows that X ∨ Y can fail to be T0 only for a pair of points in X or a pair of points in Y . But in this then case the equivalence is obvious. Next we prove (ii). Assume rst that is an open cover of i ∈ I. Then Y {Bi }i∈I Y is not compact and that without a nite subcover. Let is an open cover of For the other direction, assume that Y X ∨Y B i = Ai ∪ X {Ai }i∈I for each without a nite subcover. is compact and take an open cover, {Bi }i∈I , of X ∨ Y . By restriction, it induces an open cover of Y with Bi ∩ Y . But this cover has a nite subcover, {Bi ∩ Y }ni=1 , since Y n is compact. It follows that {Bi }i=1 is nite subcover of X ∨ Y since any Bi where Bi ∩ Y 6= ∅ covers X . To prove (iii), assume that x ∈ X and y ∈ Y . Since x ∈ N (y), it is clear that x and y are adjacent. If a, b ∈ X , then {a, y, b} a connected set for the same reason. In the same way {c, x, d} is connected if c, d ∈ Y . elements In fact all properties of the two proceeding propositions, except (iii) of Proposition 4, are true for general topological spaces, not only for smallestneighborhood spaces. 4.2 Decomposable spaces If Z Z = X ∨Y implies X = ∅ or Y = ∅, then the smallest-neighborhood space is called indecomposable, otherwise Z is called decomposable. Note that a locally nite decomposable smallest-neighborhood space is in fact nite. Proposition 6. If a smallest-neighborhood space Z is decomposable, then for every x, y ∈ Z there is a point z ∈ Z such that x, y ∈ AN(z). (If Z is T0 , an equivalent and more comprehensible condition is that ρZ (x, y) 6 2.) Proof. The result is given by the proof of Proposition 5, Part (iii). The converse implication is not true, as the following example shows. 70 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 Example 3. X = {a1 , a2 , b, c1 , c2 } be a set with 5 points, and equip it N (a1 ) = {a1 }, N (a2 ) = {a2 }, N (b) = {a1 , b}, N (c1 ) = {a1 , a2 , b, c1 }, and N (c2 ) = {a1 , a2 , c2 }. It is easy to see that ρ(x, y) 6 2 for any x, y ∈ X . Suppose that X were decomposable, X = A ∨ C . We would necessarily have a1 , a2 ∈ A since these points are open, and c1 , c2 ∈ C since these points are closed. But b cannot be in A since b 6∈ N (c2 ) and b cannot be in C since a2 6∈ N (b). Therefore, X is indecomposable. Let with a topology as follows: We have the following uniqueness result for the decomposition. Theorem 2. Let X be a smallest-neighborhood space. If X = Y ∨ Z and X = Ỹ ∨ Z̃ , where Y and Ỹ are indecomposable and non-empty, then Y = Ỹ and Z = Z̃ . Proof. It is sucient to prove that Y = Ỹ . If Y 6= Ỹ , we may suppose that Y r Ỹ is not empty and let p ∈ Y r Ỹ . Then Ỹ ⊂ Y , for if there were a point q in Ỹ r Y , then q ∈ Z so that CX (q) ⊂ Z , since X = Y ∨ Z . But by assumption p 6∈ Ỹ , so therefore p ∈ Z̃ . As q ∈ Ỹ and X = Ỹ ∨ Z̃ , this implies p ∈ CX (q) ⊂ Z , which is a contradiction since p ∈ Y . Dene B = Y r Ỹ , which is non-empty by assumption. Take two arbitrary points a ∈ Ỹ and b ∈ B . Note that b ∈ Z̃ . Hence a ∈ NX (b) and thus also a ∈ NY (b). It follows that NY (b) = NB (b) ∪ Ỹ . If Ỹ and B are equipped with the relative topology, we have Y = Ỹ ∨ B , so Y is decomposable contrary to the assumption. If a smallest-neighborhood space X is locally nite, then repeated use of Theorem 2 together with associativity gives the following. Corollary 1. If X is a locally nite smallest-neighborhood space, then X can be written in a unique way as X = Y1 ∨ · · · ∨ Yn where each Yi is indecomposable and non-empty. Note that if X is decomposable so that n > 1, then X is necessarily nite. While Theorem 2 is quite general, there is no universal cancellation law; if A∨X = B ∨X we cannot conclude that (which we will denote by Example 4. A ' B ), A and B are homeomorphic as the following example shows. N denote the set of natural numbers equipped with the N (n) = {i ∈ N; i > n}. For a positive integer m, let Nm = N ∩ [0, m], with the induced topology. It follows that N = Nm ∨ N for every m, but Nm is homeomorphic to Nk only if m = k . Let topology given by On the other hand, if X is locally nite, the unique decomposition of Corollary 1 implies that both a right and a left cancellation law hold. In fact, we may weaken the hypothesis slightly. Theorem 3. Let X be a smallest-neighborhood space. Suppose there are nitely many locally nite spaces Y1 , . . . , Yn so that X = Y1 ∨ · · · ∨ Yn . Then for all smallest-neighborhood spaces A and B we have 71 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 (i) (ii) X ∨ A ' X ∨ B implies A ' B , A ∨ X ' B ∨ X implies A ' B . Note that X n=1 is locally nite only if or if every Yi X (and hence itself ) is nite. Proof. Let Y opening chain in Y to be a nite sequence of pairwise distinct points (y0 , . . . , yn ) in Y such that yi+1 ∈ NY (yi ) for 0 6 i 6 n. The number n is the called the length of the open chain. We say that the chain starts in y0 . Let hY : Y → N ∪ {∞} We shall prove (i). The second claim follows by duality. be any smallest-neighborhood space. Dene an be dened by hY (y) = sup(n; there is an opening chain in X , it x ∈ X. Y of length n starting in From the construction of is straightforward to check that a nite number for any Furthermore, for any a∈A y). hA∨X (x) is we have hX∨A (a) > 1 + sup hX∨A (x) (1) x∈X since every x∈X belongs to NX∨A (a). Furthermore, sup hX∨A (x) = sup hX∨B (x), x∈X (2) x∈X hX (x) = hX∨Y (x) for any space Y and any x ∈ X . φ : X ∨ A → X ∨ B be a homeomorphism. A chain is mapped to a chain by φ, so we have the identity hX∨A (x) = hX∨B (φ(x)) for every x ∈ X ∨ A. In view of (1) and (2) this implies that since Let hX∨B (φ(a)) > 1 + sup hX∨B (x), x∈X for every a ∈ A. Hence φ(X) = X and φ(A) = B , which proves (i). 5. Applications We shall demonstrate how the tools we have developed can be used to give a simple proof of a known result in digital topology, namely the characterization of neighborhoods in Khalimsky spaces (Evako et al. [6]). We start with a consequence of Proposition 6. Corollary 2. Let p ∈ Zn be pure. Then A (p) is indecomposable. If p is closed, AN(p) = A (p) ∨ p and if p is open, AN(p) = p ∨ A (p). Proof. q ∈ A (p) then r = 2p − q Zn ⊂ Rn ) also belongs to A (p). It is readily checked that ρA (p) (q, r) = 4 if n > 2 (if n = 1, AZ (p) is not connected and the result immediate). Proposition 6 shows that A (p) is indecomposable. The decomposition of AN(p) is straightforward. For the rst property, notice that if (as vectors in 72 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 If p is pure, it is easy to explicitly describe AN(p). We have AN(p) = {x ∈ Zn ; kx − pk∞ 6 1}, 3n − 1 points. More generally, let q ∈ Z be any point with j so that p is adjacent to even coordinates and k odd coordinates. To simplify our notation, we note that there is a homeomor- Zn , build from a translation q 7→ q +v , where v ∈ 2Zn , and permun tation of coordinates, which takes q to the point q̃ = (0, . . . , 0, 1, . . . 1) ∈ Z , where there are j zeros and k ones (and j + k = n). phism of It follows that N (q̃) = [−1, 1]j × {1}k and that C (q̃) = {0}j × [0, 2]k . AN(q̃) = N (q̃) ∪ C (q̃), we see that q is adjacent to 3j + 3k − 2 points. j k Let 0j denote the point (0, . . . , 0) ∈ Z and let 1k = (1, . . . , 1) ∈ Z . It is easy to see that N (q̃) ' NZj (0j ) and it follows that N (q̃) r {q̃} is homeomorphic to AZj (0j ), which is indecomposable by Corollary 2. By a similar argument, C (q) r {q̃} is homeomorphic to AZk (1k ). Note that A (00 ) = A (10 ) = ∅. Since It is straightforward to check that in any smallest-neighborhood space X and for any x∈X we have A (x) ' (N (x) r {x}) ∨ (C (x) r {x}) and we obtain the following. Proposition 7. Let q ∈ Zn . Then AZn (q) ' AZj (0j ) ∨ AZk (1k ), where j is the number of even coordinates in q and k is the number of odd coordinates. As stated from the outset, this result is known, and serves only as an illustration of the formalism introduced. 6. Conclusion We have studied the join operator, which takes two topological spaces and combines them into a new space. This operation is interesting primarily for small spaces, viz. adjacency neighborhoods. We have seen that if we assume local niteness of the spaces involved, we can show that a space is decomposed in a unique way into indecomposable spaces, and we have given a criterion to recognize indecomposable spaces. The machinery can be used to systematically investigate local properties of digital topological spaces. Hopefully, this will lead to new insights into the nature of such spaces. 73 Proceedings of the 8th International Symposium on Mathematical Morphology, Rio de Janeiro, Brazil, Oct. 10 –13, 2007, MCT/INPE, v. 1, p. 63–74. http://urlib.net/dpi.inpe.br/ismm@80/2007/03.19.21.57 References [1] P. Alexandrov, Diskrete Räume, Mat. Sb. 2 (1937), no. 44, 501519. [2] P. Alexandrov and H. Hopf, Topologie I, Julius Springer, Berlin, 1935. [3] G. Birkho, Rings of sets, Duke Math J. 3 (1937), no. 3, 443454. [4] X. Daragon, M. Couprie, and G. Bertrand, Derived neighborhoods and frontier orders, Discrete Appl. Math. 147 (2005), no. 2-3, 227243. [5] A. V. Evako, Dimension on discrete spaces, Internat. J. Theoret. Phys. 33 (1994), no. 7, 15531567. [6] A. V. Evako, R. Kopperman, and Y. V. Mukhin, Dimensional properties of graphs and digital spaces, J. Math. Imaging Vision 6 (1996), no. 23, 109119. [7] G. T. Herman and D. Webster, A topological proof of a surface tracking algorithm, Computer Vision, Graphics, and Image Processing 23 (1983), 162177. [8] E. Khalimsky, R. Kopperman, and P. R. Meyer, Computer graphics and connected topologies on nite ordered sets, Topology Appl. 36 (1990), no. 1, 117. [9] C. O. Kiselman, Digital geometry and mathematical morphology, Uppsala University, 2004. Available at www.math.uu.se/~kiselman. [10] R. Klette, Topologies on the planar orthogonal grid, 6th International Conference on Pattern Recognition (ICPR'02), 2002, pp. 354357. [11] T. Y. Kong and A. Rosenfeld, Digital topology: Introduction and survey, Comput. Vision Graph. Image Process. 48 (1989), no. 3, 357393. [12] R. Kopperman, The Khalimsky line as a foundation for digital topology, Shape in picture, 1994, pp. 320. [13] V. A. Kovalevsky, Finite topology as applied to image analysis., Comput. Vision Graph. Image Process. 46 (1989), 141161. [14] E. Melin, Continuous extension in topological digital spaces, Technical Report 2004:2, Uppsala University, 2004. To appear in Applied General Topology. [15] , Continuous digitization in Khalimsky spaces, Technical Report 2006:3, Uppsala University, 2006. To appear in J. Approx. Theory. 74